PeopleofSuperintelligence

Researcher

Nate Kushman

Nate Kushman is a research scientist at Google DeepMind in London whose work connects program synthesis, mathematical reasoning, and AI for code. At MIT, he led a 2014 paper on translating algebra word problems into equations and completed a 2015 PhD with Regina Barzilay as one of his thesis advisors. By April 2017, he was a researcher at Microsoft Research Cambridge; he coauthored TerpreT and Neural Program Lattices there. His later work included PixelDefend for adversarial images, a 2020 method for robust multi-view representations, AlphaCode, and studies of feedback for mathematical reasoning. He contributed to Google DeepMind teams behind Gemini Deep Think's gold-medal-level results at the 2025 IMO and ICPC; his Deep Think team's work also underpinned a Gemini-based Paper Assistant Tool for STOC 2026.

Works
43
Citations
5,056
h-index
23
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2007

    R-BGP for Internet resilience

    In 2007, he coauthors R-BGP, which precomputes failover paths so Internet domains stay connected after link failures.

  2. 2014–2015

    Algebra problems and an MIT PhD

    In 2014, he leads a paper on solving algebra word problems by translating natural language into formal equations. In 2015, he completes an MIT PhD with Regina Barzilay as one of his thesis advisors and writes a dissertation on generating computer programs from natural-language descriptions.

  3. 2016–2017

    TerpreT and Neural Program Lattices

    By April 2017, he is a researcher at Microsoft Research Cambridge in its Machine Intelligence and Perception group. He coauthors TerpreT in 2016, a domain-specific language for expressing program synthesis problems. Neural Program Lattices uses a neural network to compose low-level programs into high-level programs in 2017.

  4. 2018

    PixelDefend against adversarial images

    In 2018, he coauthors PixelDefend, which purifies adversarial images by moving them toward the training-data distribution before an unmodified classifier processes them.

  5. 2020

    A multi-view information bottleneck

    In 2020, he coauthors Learning Robust Representations via Multi-View Information Bottleneck, which extends representation learning to an unsupervised setting with paired views and removes information that the views do not share.

  6. 2022

    AlphaCode and math feedback

    In 2022, he coauthors AlphaCode, which achieves an average rank in the top 54.3% of Codeforces contests with more than 5,000 participants. A second 2022 study of math word problems finds that correct reasoning steps need process-based supervision or learned reward models that emulate it.

  7. 2025

    Gemini at the IMO and ICPC

    In July 2025, he is named among the leads and key contributors of Google DeepMind's IMO team, whose Gemini Deep Think model solves five of six problems and earns 35 points, a gold-medal-level result. In September 2025, he contributes foundational research for Gemini 2.5 Deep Think's ICPC World Finals run, which solves 10 of 12 problems and would rank second among university teams.

  8. As of October 2026

    Research scientist at DeepMind

    He is a research scientist at Google DeepMind in London as of October 2026. For STOC 2026, a Gemini-based Paper Assistant Tool provides theoretical computer scientists with automated feedback on submission papers, building on Deep Think team work that includes him.

Career

  1. Since 2022Google DeepMindResearch Scientist
  2. 2018Microsoft ResearchResearcher
  3. 2016–2018Microsoft

Education

  1. 2015Massachusetts Institute of TechnologyPhD

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Competition-level code generation with AlphaCode2022 · Science · 1,034 citationsWith Cyprien de Masson d’Autume, Julian Schrittwieser, Koray Kavukcuoglu, Nando de Freitas, Oriol Vinyals and Po-Sen Huang
  2. Gemini: A Family of Highly Capable Multimodal Models2023 · 988 citationsWith Angeliki Lazaridou, Anja Hauth, Clemens Meyer, David Silver, Emily Pitler and Fabio Viola
  3. PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples2017 · 562 citationsWith Sebastian Nowozin
  4. Learning to Automatically Solve Algebra Word Problems2014 · Annual Meeting of the Association for Computational Linguistics (ACL) · 352 citationsWith Luke Zettlemoyer and Regina Barzilay
  5. Learning to Solve Arithmetic Word Problems with Verb Categorization2014 · Conference on Empirical Methods in Natural Language Processing (EMNLP) · 298 citationsWith Oren Etzioni

From OpenAlex, as of 21 October 2026.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Nate Kushman. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Nate Kushman. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  3. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  4. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Nate Kushman yet. Send us a free-licensed photo.

  1. R-BGP: Staying Connected in a Connected World microsoft.com. Checked 21 October 2026.
  2. Computer system automatically solves word problems news.mit.edu. Checked 21 October 2026.
  3. Nate Kushman Home Page kushman.org. Checked 21 October 2026.
  4. TerpreT: A Probabilistic Programming Language for Program Induction microsoft.com. Checked 21 October 2026.
  5. Neural Program Lattices microsoft.com. Checked 21 October 2026.
  6. Isabelle Lorge isabellelorge.github.io. Checked 21 October 2026.
  7. PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples microsoft.com. Checked 21 October 2026.
  8. Learning Robust Representations via Multi-View Information Bottleneck arxiv.org. Checked 21 October 2026.
  9. Competition-Level Code Generation with AlphaCode arxiv.org. Checked 21 October 2026.
  10. Solving math word problems with process- and outcome-based feedback arxiv.org. Checked 21 October 2026.
  11. Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad deepmind.google. Checked 21 October 2026.
  12. Gemini achieves gold-medal level at the International Collegiate Programming Contest World Finals deepmind.google. Checked 21 October 2026.
  13. Gemini-backed Paper Assistant Tool provides automated feedback for theoretical computer scientists at STOC 2026 research.google. Checked 21 October 2026.